Topic: AMD Bulldozer and SVP

good morning (excuse for my english)

I just bought an AMD FX-8150 to exploit the potential of SVP

I just finished installing Windows (32bit) with all updates:

Video drivers are the 11.10 (for video card 6670)  and have the latest DirectX (with WebUpdate)
the mpc hc is the latest stable version (1.5.3.3819)
potplayer and is version 1.5.29996
fddshow is version 4019 (compiler MSVC2010)
avisynth is your version and svp and is the 3.05

I followed exactly the procedures described by you but I have the errors in the attached files.

I did the test svpmark and I posted the result in the database to give the opportunity to others to understand the potentiality of AMD bulldozers

i think that the most important thing is understand if it is stable and compatible

from what I've seen i said that it is a bit less powerfull than i7 2600

I would be grateful if you help me to understand these errors and I remain available to do all the tests you want

thanks

Post's attachments

mpc hc error.png 376.6 kb, 262 downloads since 2011-11-10 

potplayer error.png 349.18 kb, 245 downloads since 2011-11-10 

Re: AMD Bulldozer and SVP

if can be helpful i attach the last script though the errors there are also with other settings

Post's attachments

ffdshow.avs 1.89 kb, 546 downloads since 2011-11-10 

  • MAG79
  • SVP developer
  • Offline

Re: AMD Bulldozer and SVP

Hello, feelingblue
Thank you for SVPmark results.

I have hear about this error with PotPlayer. The same error:

Unhandled exception occurred[0xE06D7363@0x74E0B9BC] at KERNELBASE.dll
Additional exception information has been stored locally and this application will be terminated.

has registered by user farookh1. (post in russian).

He resolved the problem by changing MKV splitter to *ffdShow (Haali Matroska).

About C++ error. Did you install VCRedist2008 from SVP 3.0.5 package?

Re: AMD Bulldozer and SVP

the VCRredist2008 is installed and up to date

in the potplayer i tried to use both splitters (one at a time): haali media splitter and ffdshow forced as external filter in the source/splitter section.
in the mpchc i tried to use both splitters (internal or haali)

after three hours of experiments the conclusion is:

potplayer: everything works perfectly except "the motion estmation accuracy at quarter pixel" and/or "blocks overlapping at 0.50 of block" (these two settings generates always the kernelbase error)

mpchc: everything works perfectly except "the motion extimation accuracy at quarter pixel" (this setting generate always the runtime error)

actually I have to say that my eyes do not notice the changes in most of the combinations of settings, the only things visible are:
1) the shader 13 generates less evident artifacts.
2) "bad area artifacts suppression" at "high" really eliminates artifacts but loose smootness
3) "Framerate changes" from 24-60 1.5 to 24-60 2.5x really improve smootnes (increasing artefacts)

however, the final result in terms of fluidity comparison with TV is fantastic

for me now the most important question is:
according to your experience is possible that the cause of errors may be Bulldozer or the ATI video card?
for a few day i have the possibility to replace Bulldozer with an i7 2600

I appreciate your work and availability (I am creating a paypal account to support you)

thanks

  • Rimsky
  • Installer Master
  • Offline

Re: AMD Bulldozer and SVP

feelingblue wrote:

"the motion estmation accuracy at quarter pixel" and/or "blocks overlapping at 0.50 of block" (these two settings generates always the kernelbase error)

in the "Performance graph" look at memory usage, I suppose memory usage close to 2GB limit (about 1.9-2.0Gb)

Re: AMD Bulldozer and SVP

feelingblue wrote:

the motion estmation accuracy at quarter pixel

It's a danger value - it's very easy to exceed 2 GB memory limit especially with many cores CPU.

feelingblue wrote:

is possible that the cause of errors may be Bulldozer or the ATI video card

Definitely no.

Re: AMD Bulldozer and SVP

finally i have found the problem!!

the memory usage is not the cause

i have also  tried lavsplitter with combination of various video decoders but nothing..

i have installed the same programs (SVP, C++2008, avisynth, ffdshow etc) on my other computer with phenom dual core and an nvidia 8600 and all things goes perfectly without errors (the performances are very low with 1080p)

this old PC is very "messed up" and not up to date (windows vista)

then i returned to my new PC and i disable the GPU acceleration ( the video card is the only important different element) and all works fine without errors

In real time if i activate GPU on SVP Manager the Potplayer immediately crash with famous kernelbase error (only if there are activated motion estimation accuracy or blocks overlapping)
now, with GPU on, potplayer crash also with shader23 and motion extimation at half pixel.

my last try will be to roll back to 11.9 and if errors persist i definitely will buy a nvidia 560

thanks for your advice and availability that allowed me to understand the problem

Re: AMD Bulldozer and SVP

feelingblue wrote:

the memory usage is not the cause

I'm pretty sure you're wrong wink
Memory consumption depends on:
0. frame size
1. half / quarter pixel
2. number of calculation threads -> and in GPU mode this number is almost two times larger by default.

So 8-cores CPU + quarter pixel + FullHD frame size leads to too high memory usage in GPU mode.

feelingblue wrote:

now, with GPU on, potplayer crash also with shader23 and motion extimation at half pixel

something's really wrong with your setup  hmm

Re: AMD Bulldozer and SVP

I realize that I'm just a rookie sad

you are right

according to your reply i change "number of threads" from auto to 9 (same value of i7) and now all works fine

probably with "auto" the number of threads is too high and the sum of frame size, h/q pixel, number of c.t. brings the system out of memory

now i have found very good settings with very very good video quality (with very few artefacts)

the "icing on the cake" will be the bad area artifacts suppression @ "high" with the smootness of "normal"

thanks again